Related Experiment Video
Updated: Aug 9, 2025
08:42
Design and Implementation of an fMRI Study Examining Thought Suppression in Young Women with, and At-risk, for Depression
Published on: May 19, 2015
10.8K
Identifying suicide attempts, ideation, and non-ideation in major depressive disorder from structural MRI data using
Jinlong Hu1, Yangmin Huang1, Xiaojing Zhang2
1Guangdong Key Lab of Communication and Computer Network, School of Computer Science and Engineering, South China University of Technology, Guangzhou, China.
Asian Journal of Psychiatry
|February 15, 2023
Summary
This study uses deep learning on brain scans to identify suicide risks in major depressive disorder (MDD) patients. Machine learning models pinpointed key brain regions associated with suicide attempts and ideation.
Area of Science:
- Neuroimaging
- Psychiatry
- Artificial Intelligence
Background:
- Major depressive disorder (MDD) is a significant public health concern.
- Identifying individuals at high risk for suicide attempts is crucial for timely intervention.
- Structural magnetic resonance imaging (sMRI) offers insights into brain morphology relevant to psychiatric conditions.
Purpose of the Study:
- To develop and validate deep learning models for predicting suicide risk in MDD patients using sMRI data.
- To identify specific neuroanatomical features associated with suicidal ideation (SI) and suicide attempts (SA).
- To enhance the interpretability of deep learning models in clinical neuroscience.
Main Methods:
- Collected sMRI data from 288 MDD patients categorized into suicide attempt (SA), suicidal ideation (SI), and no suicidal ideation/attempts (NS) groups.
- Developed interpretable deep neural network models for three classification tasks: SA vs. SI, SA vs. NS, and SI vs. NS.
- Employed feature extraction techniques to identify salient brain regions and features contributing to model predictions.
Main Results:
- The deep learning models successfully classified MDD patients based on suicide risk levels.
- Key structural MRI features and brain regions contributing to the classification of SA and SI were identified.
- Model interpretability allowed for the pinpointing of neuroanatomical correlates of suicidal behavior.
Conclusions:
- Deep learning analysis of sMRI data can effectively identify suicide risks in MDD patients.
- The identified neuroanatomical features provide potential biomarkers for suicide risk assessment.
- Interpretable AI models offer valuable tools for understanding the neural underpinnings of suicidal behavior in MDD.
